ap_rllib_experiment_analysis/notebooks/youshallnotpass_rllib-analysis-sizes.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from ray.tune.analysis import Analysis\n",
    "from matplotlib import pyplot as plt\n",
    "import plotly.express as px\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [],
   "source": [
    "analysis = Analysis(\"/home/sergei/ray_results/adversarial_sizes\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [],
   "source": [
    "target = 'policy_reward_mean/player_1'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = analysis.dataframe(metric=target, mode='max')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "episode_reward_max                                                          6.73144\n",
       "episode_reward_min                                                         -4.17688\n",
       "episode_reward_mean                                                         5.09732\n",
       "episode_len_mean                                                            169.154\n",
       "episodes_this_iter                                                              273\n",
       "                                                        ...                        \n",
       "config/rollout_fragment_length                                                 2866\n",
       "config/sgd_minibatch_size                                                     22628\n",
       "config/train_batch_size                                                       42880\n",
       "config/use_gae                                                                 True\n",
       "logdir                            /home/sergei/ray_results/adversarial_sizes/tra...\n",
       "Name: 0, Length: 97, dtype: object"
      ]
     },
     "execution_count": 76,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.iloc[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['episode_reward_max', 'episode_reward_min', 'episode_reward_mean',\n",
       "       'episode_len_mean', 'episodes_this_iter', 'num_healthy_workers',\n",
       "       'timesteps_total', 'done', 'episodes_total', 'training_iteration',\n",
       "       'experiment_id', 'date', 'timestamp', 'time_this_iter_s',\n",
       "       'time_total_s', 'pid', 'hostname', 'node_ip', 'time_since_restore',\n",
       "       'timesteps_since_restore', 'iterations_since_restore',\n",
       "       'checkpoint_rllib', 'trainer_iteration', 'trial_id', 'experiment_tag',\n",
       "       'policy_reward_min/player_1', 'policy_reward_min/player_2',\n",
       "       'policy_reward_max/player_1', 'policy_reward_max/player_2',\n",
       "       'policy_reward_mean/player_1', 'policy_reward_mean/player_2',\n",
       "       'hist_stats/episode_reward', 'hist_stats/episode_lengths',\n",
       "       'hist_stats/policy_player_1_reward',\n",
       "       'hist_stats/policy_player_2_reward', 'sampler_perf/mean_env_wait_ms',\n",
       "       'sampler_perf/mean_processing_ms', 'sampler_perf/mean_inference_ms',\n",
       "       'timers/sample_time_ms', 'timers/sample_throughput',\n",
       "       'timers/learn_time_ms', 'timers/learn_throughput',\n",
       "       'timers/update_time_ms', 'info/num_steps_sampled',\n",
       "       'info/num_steps_trained', 'perf/cpu_util_percent',\n",
       "       'perf/ram_util_percent', 'perf/gpu_util_percent0',\n",
       "       'perf/vram_util_percent0', 'perf/gpu_util_percent1',\n",
       "       'perf/vram_util_percent1', 'perf/gpu_util_percent2',\n",
       "       'perf/vram_util_percent2', 'perf/gpu_util_percent3',\n",
       "       'perf/vram_util_percent3', 'info/learner/player_1/cur_kl_coeff',\n",
       "       'info/learner/player_1/cur_lr', 'info/learner/player_1/total_loss',\n",
       "       'info/learner/player_1/policy_loss', 'info/learner/player_1/vf_loss',\n",
       "       'info/learner/player_1/vf_explained_var', 'info/learner/player_1/kl',\n",
       "       'info/learner/player_1/entropy', 'info/learner/player_1/entropy_coeff',\n",
       "       'config/_base_dir', 'config/_call', 'config/_env', 'config/_env_fcn',\n",
       "       'config/_env_name_rllib', 'config/_get_policies', 'config/_log_error',\n",
       "       'config/_main_filename', 'config/_model_params',\n",
       "       'config/_num_workers_tf', 'config/_policies', 'config/_policy',\n",
       "       'config/_redis_address', 'config/_run_inline', 'config/_select_policy',\n",
       "       'config/_tmp_dir', 'config/_train_policies', 'config/_train_steps',\n",
       "       'config/_trainer', 'config/_update_config', 'config/batch_mode',\n",
       "       'config/framework', 'config/kl_coeff', 'config/lr',\n",
       "       'config/num_envs_per_worker', 'config/num_gpus', 'config/num_sgd_iter',\n",
       "       'config/num_workers', 'config/rollout_fragment_length',\n",
       "       'config/sgd_minibatch_size', 'config/train_batch_size',\n",
       "       'config/use_gae', 'logdir'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 77,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = df.sort_values(target, ascending=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([3., 2., 0., 0., 3., 0., 2., 1., 0., 1.]),\n",
       " array([-7.06959707, -6.34939233, -5.62918759, -4.90898285, -4.18877811,\n",
       "        -3.46857337, -2.74836863, -2.02816389, -1.30795915, -0.58775441,\n",
       "         0.13245033]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 79,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(df[target])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [],
   "source": [
    "df['fcnet_hiddens'] = [x['fcnet_hiddens'] for x in df['config/_model_params']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [],
   "source": [
    "def weights(arr):\n",
    "    arr = [380] + list(arr) + [17]\n",
    "    x_old = arr[0]\n",
    "    w = 0\n",
    "    for x in arr[1:]:\n",
    "        w += x_old * x + x\n",
    "        x_old = x\n",
    "    return w\n",
    "df['layers'] = [len(x) for x in df['fcnet_hiddens']]\n",
    "df['neurons'] = [x[0] for x in df['fcnet_hiddens']]\n",
    "df['weights'] = [weights(x) for x in df['fcnet_hiddens']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.scatter(df['weights'], df[target])\n",
    "plt.xscale('log')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.collections.PathCollection at 0x7f65e3a85e90>"
      ]
     },
     "execution_count": 100,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.scatter(df['layers'], df[target])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.collections.PathCollection at 0x7f65e398ded0>"
      ]
     },
     "execution_count": 101,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.scatter(df['neurons'], df[target])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "policy_reward_mean/player_1                                                -3.64121\n",
       "config/_base_dir                        /scratch/sergei/better-adversarial-defenses\n",
       "config/_call                      {'checkpoint_freq': 0, 'name': 'adversarial_li...\n",
       "config/_env                       {'env_name': 'multicomp/YouShallNotPassHumans-...\n",
       "config/_env_fcn                             <function create_env at 0x7f6462e5b710>\n",
       "config/_env_name_rllib                                                    multicomp\n",
       "config/_log_error                                                              True\n",
       "config/_main_filename             /scratch/sergei/better-adversarial-defenses/tr...\n",
       "config/_model_params              {'custom_model': 'LinearModel', 'custom_model_...\n",
       "config/_num_workers_tf                                                            4\n",
       "config/_policies                                   [None, from_scratch, pretrained]\n",
       "config/_policy                                                                  PPO\n",
       "config/_redis_address                                            128.32.175.10:6379\n",
       "config/_run_inline                                                            False\n",
       "config/_tmp_dir                                                                /tmp\n",
       "config/_train_policies                                                   [player_1]\n",
       "config/_train_steps                                                        99999999\n",
       "config/_trainer                                                                 PPO\n",
       "config/_update_config                                                          None\n",
       "config/batch_mode                                                 complete_episodes\n",
       "config/framework                                                                tfe\n",
       "config/kl_coeff                                                                   1\n",
       "config/lr                                                               0.000755454\n",
       "config/num_envs_per_worker                                                        4\n",
       "config/num_gpus                                                                   0\n",
       "config/num_sgd_iter                                                               5\n",
       "config/num_workers                                                                8\n",
       "config/rollout_fragment_length                                                 2866\n",
       "config/sgd_minibatch_size                                                     22628\n",
       "config/train_batch_size                                                       42880\n",
       "config/use_gae                                                                 True\n",
       "logdir                            /home/sergei/ray_results/adversarial_linear/tr...\n",
       "Name: 3, dtype: object"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best_trial = df.iloc[2].logdir\n",
    "df.iloc[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>policy_reward_mean/player_1</th>\n",
       "      <th>config/_base_dir</th>\n",
       "      <th>config/_call</th>\n",
       "      <th>config/_env</th>\n",
       "      <th>config/_env_fcn</th>\n",
       "      <th>config/_env_name_rllib</th>\n",
       "      <th>config/_log_error</th>\n",
       "      <th>config/_main_filename</th>\n",
       "      <th>config/_model_params</th>\n",
       "      <th>config/_num_workers_tf</th>\n",
       "      <th>...</th>\n",
       "      <th>config/lr</th>\n",
       "      <th>config/num_envs_per_worker</th>\n",
       "      <th>config/num_gpus</th>\n",
       "      <th>config/num_sgd_iter</th>\n",
       "      <th>config/num_workers</th>\n",
       "      <th>config/rollout_fragment_length</th>\n",
       "      <th>config/sgd_minibatch_size</th>\n",
       "      <th>config/train_batch_size</th>\n",
       "      <th>config/use_gae</th>\n",
       "      <th>logdir</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>-3.641208</td>\n",
       "      <td>/scratch/sergei/better-adversarial-defenses</td>\n",
       "      <td>{'checkpoint_freq': 0, 'name': 'adversarial_li...</td>\n",
       "      <td>{'env_name': 'multicomp/YouShallNotPassHumans-...</td>\n",
       "      <td>&lt;function create_env at 0x7f6462e5b710&gt;</td>\n",
       "      <td>multicomp</td>\n",
       "      <td>True</td>\n",
       "      <td>/scratch/sergei/better-adversarial-defenses/tr...</td>\n",
       "      <td>{'custom_model': 'LinearModel', 'custom_model_...</td>\n",
       "      <td>4</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000755</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>8</td>\n",
       "      <td>2866</td>\n",
       "      <td>22628</td>\n",
       "      <td>42880</td>\n",
       "      <td>True</td>\n",
       "      <td>/home/sergei/ray_results/adversarial_linear/tr...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>-5.884354</td>\n",
       "      <td>/scratch/sergei/better-adversarial-defenses</td>\n",
       "      <td>{'checkpoint_freq': 0, 'name': 'adversarial_li...</td>\n",
       "      <td>{'env_name': 'multicomp/YouShallNotPassHumans-...</td>\n",
       "      <td>&lt;function create_env at 0x7f6462e5b710&gt;</td>\n",
       "      <td>multicomp</td>\n",
       "      <td>True</td>\n",
       "      <td>/scratch/sergei/better-adversarial-defenses/tr...</td>\n",
       "      <td>{'custom_model': 'LinearModel', 'custom_model_...</td>\n",
       "      <td>4</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000755</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>8</td>\n",
       "      <td>2866</td>\n",
       "      <td>22628</td>\n",
       "      <td>42880</td>\n",
       "      <td>True</td>\n",
       "      <td>/home/sergei/ray_results/adversarial_linear/tr...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>-8.736462</td>\n",
       "      <td>/scratch/sergei/better-adversarial-defenses</td>\n",
       "      <td>{'checkpoint_freq': 0, 'name': 'adversarial_li...</td>\n",
       "      <td>{'env_name': 'multicomp/YouShallNotPassHumans-...</td>\n",
       "      <td>&lt;function create_env at 0x7f6462e5b710&gt;</td>\n",
       "      <td>multicomp</td>\n",
       "      <td>True</td>\n",
       "      <td>/scratch/sergei/better-adversarial-defenses/tr...</td>\n",
       "      <td>{'custom_model': 'LinearModel', 'custom_model_...</td>\n",
       "      <td>4</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000755</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>8</td>\n",
       "      <td>2866</td>\n",
       "      <td>22628</td>\n",
       "      <td>42880</td>\n",
       "      <td>True</td>\n",
       "      <td>/home/sergei/ray_results/adversarial_linear/tr...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>-9.108062</td>\n",
       "      <td>/scratch/sergei/better-adversarial-defenses</td>\n",
       "      <td>{'checkpoint_freq': 0, 'name': 'adversarial_li...</td>\n",
       "      <td>{'env_name': 'multicomp/YouShallNotPassHumans-...</td>\n",
       "      <td>&lt;function create_env at 0x7f6462e5b710&gt;</td>\n",
       "      <td>multicomp</td>\n",
       "      <td>True</td>\n",
       "      <td>/scratch/sergei/better-adversarial-defenses/tr...</td>\n",
       "      <td>{'custom_model': 'LinearModel', 'custom_model_...</td>\n",
       "      <td>4</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000755</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>8</td>\n",
       "      <td>2866</td>\n",
       "      <td>22628</td>\n",
       "      <td>42880</td>\n",
       "      <td>True</td>\n",
       "      <td>/home/sergei/ray_results/adversarial_linear/tr...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>4 rows × 32 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   policy_reward_mean/player_1                             config/_base_dir  \\\n",
       "3                    -3.641208  /scratch/sergei/better-adversarial-defenses   \n",
       "0                    -5.884354  /scratch/sergei/better-adversarial-defenses   \n",
       "2                    -8.736462  /scratch/sergei/better-adversarial-defenses   \n",
       "1                    -9.108062  /scratch/sergei/better-adversarial-defenses   \n",
       "\n",
       "                                        config/_call  \\\n",
       "3  {'checkpoint_freq': 0, 'name': 'adversarial_li...   \n",
       "0  {'checkpoint_freq': 0, 'name': 'adversarial_li...   \n",
       "2  {'checkpoint_freq': 0, 'name': 'adversarial_li...   \n",
       "1  {'checkpoint_freq': 0, 'name': 'adversarial_li...   \n",
       "\n",
       "                                         config/_env  \\\n",
       "3  {'env_name': 'multicomp/YouShallNotPassHumans-...   \n",
       "0  {'env_name': 'multicomp/YouShallNotPassHumans-...   \n",
       "2  {'env_name': 'multicomp/YouShallNotPassHumans-...   \n",
       "1  {'env_name': 'multicomp/YouShallNotPassHumans-...   \n",
       "\n",
       "                           config/_env_fcn config/_env_name_rllib  \\\n",
       "3  <function create_env at 0x7f6462e5b710>              multicomp   \n",
       "0  <function create_env at 0x7f6462e5b710>              multicomp   \n",
       "2  <function create_env at 0x7f6462e5b710>              multicomp   \n",
       "1  <function create_env at 0x7f6462e5b710>              multicomp   \n",
       "\n",
       "   config/_log_error                              config/_main_filename  \\\n",
       "3               True  /scratch/sergei/better-adversarial-defenses/tr...   \n",
       "0               True  /scratch/sergei/better-adversarial-defenses/tr...   \n",
       "2               True  /scratch/sergei/better-adversarial-defenses/tr...   \n",
       "1               True  /scratch/sergei/better-adversarial-defenses/tr...   \n",
       "\n",
       "                                config/_model_params  config/_num_workers_tf  \\\n",
       "3  {'custom_model': 'LinearModel', 'custom_model_...                       4   \n",
       "0  {'custom_model': 'LinearModel', 'custom_model_...                       4   \n",
       "2  {'custom_model': 'LinearModel', 'custom_model_...                       4   \n",
       "1  {'custom_model': 'LinearModel', 'custom_model_...                       4   \n",
       "\n",
       "   ... config/lr config/num_envs_per_worker config/num_gpus  \\\n",
       "3  ...  0.000755                          4               0   \n",
       "0  ...  0.000755                          4               0   \n",
       "2  ...  0.000755                          4               0   \n",
       "1  ...  0.000755                          4               0   \n",
       "\n",
       "   config/num_sgd_iter config/num_workers config/rollout_fragment_length  \\\n",
       "3                    5                  8                           2866   \n",
       "0                    5                  8                           2866   \n",
       "2                    5                  8                           2866   \n",
       "1                    5                  8                           2866   \n",
       "\n",
       "   config/sgd_minibatch_size config/train_batch_size config/use_gae  \\\n",
       "3                      22628                   42880           True   \n",
       "0                      22628                   42880           True   \n",
       "2                      22628                   42880           True   \n",
       "1                      22628                   42880           True   \n",
       "\n",
       "                                              logdir  \n",
       "3  /home/sergei/ray_results/adversarial_linear/tr...  \n",
       "0  /home/sergei/ray_results/adversarial_linear/tr...  \n",
       "2  /home/sergei/ray_results/adversarial_linear/tr...  \n",
       "1  /home/sergei/ray_results/adversarial_linear/tr...  \n",
       "\n",
       "[4 rows x 32 columns]"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "957\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "best_df = analysis.trial_dataframes[best_trial]\n",
    "rew = best_df[target]\n",
    "plt.plot(rew)\n",
    "print(np.argmax(rew))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'/home/sergei/ray_results/adversarial_linear/train_one_with_sacred_0_2020-08-24_20-10-37qqj3lwd8'"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best_trial"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'/home/sergei/ray_results/PPO_multicomp_2020-08-25_10-42-17yx9c3k0s/checkpoint_387/checkpoint-387'"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best_df.checkpoint_rllib.iloc[-1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}